Deadlock in NAT detect

[nanang]

It is caused by non-uniform lock ordering between NAT detect and STUN session.

Scenario:

Thread 1:

1. NAT detect on_request_complete() is called by STUN session while holding STUN session lock (the lock is a loose/independent group lock)
2. in the callback, NAT detect is about to acquire its own lock

Thread 2:

1. NAT detect on_read_complete() is called by ioqueue, and soon NAT detect lock is acquired
2. in the callback, NAT detect calls pj_stun_session_on_rx_pkt() which tries to acquire STUN session lock

[nanang]

In 4573:

Fix #1691: Apply group lock mechanism in NAT detect to avoid deadlock.